1. Field of the Invention
The present invention relates to a capable semiconductor device such as a micromachine, a solar battery having a high energy conversion efficiency, a volatile or a nonvolatile memory, a load of a memory cell of an SRAM, a field effective thin film transistor (hereinafter referred to as "TFT") which is used as a display device of a wide-screen, high-precision active-matrix type liquid crystal display and a liquid crystal display which comprises a pixel electrode and a thin film transistor. The present invention also relates to a method for manufacturing such a semiconductor device.
2. Background of the Invention
(Prior Art 1)
A prototype of a micromachine, which is typically an apparatus of fine dimensions in the units of micrometers or less having a mechanical motion mechanism, is prepared by redeployment of a semiconductor technique mainly implementing simultaneous formation on a semiconductor wafer by transfer of a mask pattern or the like.
&lt;Micromachine Having Movable Part&gt;
FIG. 77 is a front sectional view showing a sectional structure of an electrostatic motor 3000 which is manufactured by a conventional method. In this electrostatic motor 3000, a nitride film (Si.sub.3 N.sub.4 film) 3004 is formed on a single-crystalline Si substrate 3002, and a rotary shaft 3014 and a stator 3010 of polycrystalline Si are fixedly formed on the nitride film 3004, while a ring-shaped rotator 3008 of polycrystalline Si is stopped to the rotary shaft 3014 with a clearance. Thus, the rotator 3008 is rotatable about the rotary shaft 3014.
The stator 3010 is formed by a plurality of members, being electrically isolated from each other, which are radially arranged at prescribed intervals to enclose the rotator 3008. Alternating voltages which are out of phase with each other are independently applied to these members of the stator 3010. Consequently, electrostatic attractive or repulsive force is generated between the stator 3010 and the rotator 3008, and a bearer of such force is successively moved between the respective members, thereby rotating the rotator 3008. It is reported that the rotator 3008 is 100 .mu.m in diameter and 2.5 .mu.m in thickness.
FIG. 78 is a process diagram showing a certain stage in a method of manufacturing the electrostatic motor 3000. In manufacturing of the electrostatic motor 3000, a technique of forming sacrifice layers and etching the same is employed in order to isolate the rotator 3008, the stator 3010, the rotary shaft 3014 and the nitride film 3004 serving as an underlayer from each other and defining clearances therebetween. In order to manufacture the electrostatic motor 3000, an SiO.sub.2 layer is first temporarily formed on an upper surface of the nitride film 3004 and selectively removed to form a first sacrifice layer 3006, for selectively filling up spaces corresponding to clearances. A polycrystalline Si layer is stacked thereon by CVD (chemical vapor deposition) or the like. Thereafter this polycrystalline Si layer is selectively removed to be formed into the shapes of the rotator 3008 and the stator 3010. Another sacrifice layer 3012 is newly formed so that another polycrystalline Si layer is newly stacked thereon, whereafter this polycrystalline Si layer is selectively removed to be formed into the shape of the rotary shaft 3014. Finally, the two sacrifice layers 3006 and 3012 are entirely removed by etching, thereby completing the electrostatic motor 3000 shown in FIG. 77.
&lt;Micromachine Having Deformed Part&gt;
FIG. 79 is a plan view showing an electrostatic linear actuator 3030 which is manufactured by a conventional method. In this actuator 3030, fixed electrodes 3040 of polycrystalline Si and movable parts 3036 of polycrystalline Si are formed on a single-crystalline Si substrate 3032. The comb-shaped fixed electrodes 3040 are fixed onto the single-crystalline Si substrate 3032 in base portions thereof, so that comb-shaped protrusions are upwardly raised from the single-crystalline Si substrate 3032. On the other hand, the movable parts 3036 are fixed to the single-crystalline Si substrate 3032 only in support portions 3034, so that other portions are entirely upwardly raised from the single-crystalline Si substrate 3032. The movable parts 3036 have comb-shaped forward end portions 3038, which are fitted with the fixed electrodes 3040 with no contact.
When voltages are applied to the fixed electrodes 3040, electrostatic attractive or repulsive force acts across the same and the forward end portions 3038, whereby the forward end portions 3038 are straightly displaced in a horizontal plane. At this time, members coupling the forward end portions 3038 with the support portions 3034, which are elastic members serving as sources of restoring force for the displacement of the forward end portions 3038, are elastically deformed following the displacement of the forward end portions 3038.
A method of manufacturing the actuator 3030 also employs a step of forming a sacrifice layer and removing the same by etching, similarly to the method of manufacturing the electrostatic motor 3000. Namely, a sacrifice layer is first formed on the single-crystalline Si substrate 3032, so that a polycrystalline Si layer is formed thereon. Thereafter the polycrystalline Si layer is selectively removed, to be formed into the shapes of the movable parts 3036 and the fixed electrodes 3040. Finally, the sacrifice layer is entirely removed by etching, to complete the actuator 3030.
&lt;Micromachine Having Cavity&gt;
FIG. 80 is a fragmented perspective view showing a part of a pressure sensor 3100 which is manufactured by a conventional method. In this pressure sensor 3100, a lower housing 3102 and an upper housing 3104 having grooves are joined with each other to form a hollow vessel defining a cavity 3106 in its interior. A plate-type oscillator 3108 is inserted in this cavity 3106. All of the lower housing 3102, the upper housing 3104 and the oscillator 3108 are made of single-crystalline Si.
Only an end portion of the oscillator 3108 is in contact with an inner wall portion defining the cavity 3106 to be supported by the hollow vessel, while all of the remaining portions are separated from the inner walls. Therefore, it is possible to excite natural oscillation in the oscillator 3108 by applying a stationary magnetic field and an oscillating current from the exterior. The lower housing 3102 is coupled with a diaphragm (not shown) which receives a pressure to be measured. The diaphragm is deformed upon application of the pressure, whereby the oscillator 3108 is also deformed. Upon such deformation of the oscillator 3108, the natural frequency deviates in response to the degree of the displacement. The level of the pressure is measured by detecting the deviation of the natural frequency. This pressure sensor 3100 has an excellent elastic limit, excellent strength and the like, as well as uniform characteristics since all of the members 3102, 3104 and 3108 are made of single-crystalline Si. Therefore, it is possible to provide these members with large deformation, while implementing a pressure sensor having high reliability.
In order to manufacture the pressure sensor 3100, a technique of junction (also referred to as "cladding") is employed as hereinabove described. After the two housings 3102 and 3104 having grooves and the oscillator 3108 are prepared from single-crystalline Si independently of each other, the two housings 3102 and 3104 are joined with each other at a junction plane 3110 while receiving the oscillator 3108 in the grooves. This junction step is carried out in a vacuum, whereby the cavity 3106 is maintained in a vacuum state also after completion of the apparatus.
&lt;Micromachine Having Diaphragm&gt;
FIG. 81 is a front elevational view showing another pressure sensor 3200 which is manufactured by a conventional method. Junction of single-crystalline Si is employed also in manufacturing of the pressure sensor 3200. Namely, a V-shaped groove 3204 is formed on a single-crystalline Si substrate 3202 by etching, whereafter a thin substrate 3206 of single-crystalline Si is joined thereto at a junction plane 3208. Thereafter etching is further advanced to expand the V-shaped groove 3204 to be opened in a bottom surface of the substrate 3202. The etching step is carried out with an etchant for selectively etching a specific crystal plane.
In the thin substrate 3206, a portion 3212 covering the groove 3204 serves as a diaphragm which receives a pressure to be measured. The diaphragm 3212 is elastically deformed by such application of the pressure. A piezoresistance element 3210 serving as a distortion gauge is formed in the vicinity of a portion having the maximum distortion following the elastic deformation, i.e., a fixed end of the diaphragm 3212. This piezoresistance element 3210 is formed by selectively implanting impurity ions into a specific portion of an upper surface of the diaphragm 3212. This pressure sensor 3200 is adapted to measure the value of the pressure through an amount of change in electric resistance in the piezoresistance element 3210. According to this pressure sensor 3200, large elastic deformation can be applied to the diaphragm 3212, since this diaphragm 3212 is made of single-crystalline Si. &lt;Micromachine Provided with Element on Insulating Layer&gt;
FIG. 82 is a front elevational view showing still another pressure sensor 3230 which is manufactured by a conventional method. Junction of single-crystalline Si is employed also in manufacturing of the pressure sensor 3230. Namely, an insulating film 3236 of SiO.sub.2 is first formed on a single-crystalline Si substrate 3232 having a diaphragm 3234 on a groove. Then, a single-crystalline Si thin film is joined onto this insulating film 3236. A p-type impurity is previously selectively introduced into this single-crystalline Si thin film, to form a piezoresistance element 3238 of p-type Si for serving as a distortion gauge. After completion of the junction, the single-crystalline Si thin film is etched for leaving the p-type Si portion while selectively removing the remaining portions. At this time, such a property is utilized that Si containing large amounts of an Si oxide (SiO.sub.2) and an impurity is at a lower etching rate as compared with Si containing a small amount of impurity. Consequently, only the piezoresistance element 3238 is left on the insulating film 3236 of SiO.sub.2. Thereafter a metal interconnection layer 3240 is selectively formed on the insulating film 3236, thereby completing the pressure sensor 3230.
The diaphragm 3224 is elastically deformed upon application of a pressure to be measured, whereby distortion following this elastic deformation changes electric resistance of the piezoresistance element 3238. The value of the pressure is measured through an amount of change of the electric resistance.
In this pressure sensor 3230, the piezoresistance element 3238 is formed on the insulating film 3236, to be electrically isolated from the substrate 3232. Therefore, no current flowing in the piezoresistance element 3238 leaks to the substrate 3232, whereby the amount of change of the electric resistance can be accurately measured.
&lt;Micromachine Having Cantilever&gt;
FIG. 83 is a perspective view showing an acceleration sensor 3300 which is manufactured by a conventional method. In this acceleration sensor 3300, an SiO.sub.2 film 3304 is formed on a single-crystalline Si substrate 3302 having a groove 3308, while a cantilever 3306 forming a part of the SiO.sub.2 film 3304 protrudes above the groove 3308. Elastic deformation of the cantilever 3306 is detected by a distortion gauge, to measure the value of acceleration or vibration.
In order to manufacture this acceleration sensor 3300, an SiO.sub.2 film is formed on the overall surface of the substrate 3302 before formation of the groove 3308, whereafter the SiO.sub.2 film is selectively removed to be formed into the shape of the SiO.sub.2 3304 film having a protrusion. Then, the upper surface of the substrate 3302 is selectively etched to be provided with the groove 3308, while the protrusion of the SiO.sub.2 film 3304 defines the cantilever 3306.
(Problem of Prior Art 1)
However, the conventional micromachines which are manufactured in the aforementioned manners according to prior art 1 have the following problems:
First, the electrostatic motor 3000 (FIGS. 77 and 78) is so hardly worn that the same cannot withstand practical application to a motor, since both of the rotator 3008 and the rotary shaft 3014 which are in sliding contact with each other are made of polycrystalline Si. These members are made of polycrystalline Si since there is no technique of forming single-crystalline Si on the sacrifice layers 3006 and 3012 which are made of SiO.sub.2. In the actuator 3030 (FIG. 79), on the other hand, the level of elastic deformation of the movable parts 3036 is limited due to restriction in elastic limit and strength, since the movable parts 3036 are made of polycrystalline Si. The movable parts 3036 are made of not single-crystalline but polycrystalline Si since the same are formed by stacking Si on the sacrifice layers of SiO.sub.2.
In order to implement the complicated structure of the pressure sensor 3100 (FIG. 80) having the cavity 3106 storing the oscillator 3108 in its interior with single-crystalline Si, only the junction technique is generally employable since all members 3102, 3104 and 3108 are made of single-crystalline Si. Thus, the manufacturing steps are complicated since it is necessary to prepare the respective members independently of each other and high dimensional accuracy is required for the respective members to be joined to each other, while the members cannot be easily aligned with each other in junction.
Also in the pressure sensor 3200 (FIG. 81), it is necessary to previously prepare the two members 3202 and 3206 which are made of single-crystalline Si independently of each other for thereafter combining the same with each other by the junction technique, in order to implement the complicated structure provided with the groove 3204. Therefore, the manufacturing steps are complicated and difficult.
Also in the pressure sensor 3230 (FIG. 82), it is necessary to employ the aforementioned junction technique, in order to form the piezoresistance element 3238 of single-crystalline Si on the insulating film 3236 of SiO.sub.2. Therefore, the manufacturing steps are complicated and difficult, similarly to the above.
The acceleration sensor 3300 (FIG. 83) is insufficient in elastic limit, strength, uniformity of characteristics and the like since the cantilever 3306 is made of SiO.sub.2. In order to manufacture the acceleration sensor 3300 without employing the junction technique, it is necessary to selectively remove the substrate 3302 of single-crystalline Si while leaving the cantilever 3306 by etching. In general, therefore, the cantilever 3306 must be made of SiO.sub.2 or the like, dissimilarly to the substrate 3302.
(Prior Art 2)
FIGS. 99 to 102 illustrate steps of manufacturing a solar cell according to Prior Art 2. Description is now made with reference to an exemplary case of forming CdTe and CdS films, which are main materials for the solar cell, by printing, in addition to formation of electrodes. As shown in FIG. 99, a paste-type CdS film 3402 is first screen-printed on an upper surface of a glass substrate 3401. The CdS film 3402 is then dried by irradiation with infrared rays, and thereafter heated in a furnace to about 650.degree. C., to be converted to a sintered semiconductor. Then this CdS film 3402 is doped with an impurity to be converted to an n-type film. Then, a CdTe film 3403 is selectively screen-printed on a part of an upper surface of the CdS film 3402 as shown in FIG. 100, dried and sintered. Then, a carbon electrode 3404 (accepter source) is printed on an upper surface of the CdTe film 3403, dried and baked, as shown in FIG. 101. At this time, an element which is converted to an accepter at about 300.degree. C. is diffused in the CdTe film 3403 to form a strong p-type CdTe layer, thereby forming a p-n junction. Thereafter an Ag-In electrode 3405 is printed/baked on an upwardly exposed portion of the CdS film 3402 while Ag electrodes 3406 are selectively printed/baked on parts of an upper surface of the carbon electrode 3404, as shown in FIG. 102.
While the solar cell shown in FIGS. 99 to 102 employs CdS belonging to the groups II and VI as an n-type film which is formed on a substrate, this film may alternatively be prepared from Si belonging to the group VI or GaAs belonging to the groups III and V, for example.
(Prior Art 3)
FIG. 103A illustrates a conventional solar cell of a tandem structure (stack system). Referring to FIG. 103A, numeral 3407 denotes a first semiconductor part (first heterojunction part), numeral 3408 denotes a second semiconductor part (second heterojunction part), and numeral 3409 denotes a tunnel junction part by high concentration junction. In general, a wavelength range of light which can be absorbed by a semiconductor is widened as a forbidden bandwidth Eg of the semiconductor is reduced, while the coefficient of energy use of the as-absorbed light is contradictorily deteriorated as the forbidden bandwidth is reduced. In the tandem structure, p-n junctions having different forbidden bandwidths Eg are overlappingly arranged in the direction of progress of the light along the order of the forbidden bandwidths Eg. As shown in FIG. 103B, the light is absorbed by the respective materials from a component of a shorter wavelength side, to develop voltages corresponding to Eg.sub.A and Eg.sub.B respectively. The semiconductor parts 3408 and 3409 having respective p-n junctions are connected in series, while excellent connection is attained through a tunnel current of the p.sup.+ -n.sup.+ tunnel junction part 3409 with no interposition of a metal therebetween. Referring to FIG. 103B, symbol E.sub.C denotes a lower end of a conduction band, symbol E.sub.V denotes an upper end of a valence band, and symbol E.sub.F denotes a Fermi level, serving as a scale for expressing to which energy level electrons are filled in a solid. These energy gaps are naturally determined by the materials employed for the respective semiconductor films. The material for the tandem structure may be prepared from a combination of GaAlAs and GaAs, or a combination of a-Si and a-SiGe. A number of at least two is considered for the layers. Prior Art art 3 is also formed by a method of screen printing - drying - sintering in general, similarly to Prior Art 2.
(Problem of Prior Arts 2 and 3)
In each of the solar cells according to Prior Art 2 and Prior Art 3 shown in FIGS. 99 to 102, 103A and 103B, however, a single-crystalline thin film cannot be obtained since the solar cell is formed by the method of screen printing - drying - sintering, for example, and its crystal structure is generally amorphous. Thus, it is difficult to add a p-type or n-type impurity as well as to control the amount of introduction of the impurity, due to the heterogeneous crystal structure.
When a single-crystalline semiconductor substrate is employed as an underlayer, it is possible to form a single-crystalline film on an upper surface of the single-crystalline substrate serving as a seed crystal. However, the single-crystalline substrate itself is extremely high-priced, contrarily to requirement for reduction in cost. Further, a single-crystalline film must be formed under high-temperature environment in Prior Art 2 and Prior Art 3, and hence the substrate must be prepared from that which can withstand a high temperature, leading to increase in cost. Particularly when a plurality of semiconductor parts 3407 and 3408 having different forbidden bandwidths are stacked as in Prior Art 3, it is necessary to vary the materials therefor and hence preparation of a seed crystal from the underlayer itself is impossible.
When a semiconductor film is prepared from Si, for example, an absorption coefficient of crystalline Si is only 10.sup.4 /cm even in a visible part, and a considerably thick plate-type material is required for entirely absorbing sunlight, such that a thickness of several 100 .mu.m is required for a polycrystalline material. On the other hand, an amorphous material may have a relatively small thickness of about 1 .mu.m, for example, since the same has a large absorption coefficient. However, its conversion efficiency is deteriorated to about 5 to 10%, due to low electron mobility and a short carrier life time. To this end, GaAs is suitable as a material, while the same is not suitable in practice due to a high cost. A group II-VI or III-V compound is also preferable in consideration of an absorption coefficient and wavelengths of absorption ends and hence CdS, for example, has been temporarily studied. However, this material is hardly single-crystallized, and it is difficult to add a p-type or n-type impurity thereto. If the group II-VI or III-V compound can be single-crystallized and doped with an impurity, however, this compound is rather preferable as compared with Si. Thus, awaited is a method which can easily single-crystallize a group II-VI or III-V compound.
(Prior Arts 4 to 15)
FIG. 127 shows characteristics of materials for general solar cells. As shown in FIG. 127, the solar cells are classified into Si system and compound system ones, depending on the materials as employed. Most of the generally produced solar cells are prepared from Si-system materials. The Si-system materials include crystalline and amorphous materials. The crystallin materials are classified into single-crystalline Si and polycrystalline Si materials, depending on substrates as employed. Among the above, solar cells of the single-crystalline Si and polycrystalline Si materials are generally employed as power systems.
Although a solar cell employing single-crystalline Si has high conversion efficiency of 17% (to 20%), a single-crystalline Si substrate is disadvantageously high-priced.
On the other hand, a solar cell employing polycrystalline Si can be provided at a lower cost than the single-crystalline Si substrate, although its conversion efficiency of 12 to 14% is inferior to that of the single-crystalline Si solar cell. Also in such polycrystalline Si, however, a constant thickness or cutting allowance is required for a bulk substrate as employed so far as the bulk substrate is formed by cutting an ingot, leading to a high cost for the substrate.
An amorphous Si solar cell is expected in view of reduction in cost. The amorphous Si solar cell is manufactured by stacking thin films through a method such as plasma CVD (chemical vapor deposition). According to this method, the thin films can be formed at a low temperature of 200 to 300.degree. C., whereby a low-priced solar cell can be manufactured since a low-priced glass material or the like can be employed and the amount of the Si material can be extremely reduced as compared with a case of employing single-crystalline or polycrystalline bulk silicon. In order to put this solar cell into practice on a full scale, however, it is necessary to solve a problem of how to improve conversion efficiency, which is 6 to 8% under the present circumstances. Further, the amorphous Si is deteriorated by about 15% (initial deterioration) in several months by time change of the conversion efficiency due to a large amount of crystal defects, and thereafter further deteriorated by about 1 to 2% every year.
Description is now made on directions of research and development which are made by respective makers for solving the aforementioned problems of the prior art.
1) Single-Crystalline Si Solar Cell (Prior Art 4)
As shown in FIG. 128A, a polycrystalline base material 3501 is prepared by trichlorosilane which is lifted by the Czochralski method (CZ method), i.e., reduced with hydrogen with application of thermal energy, and single-crystallized with application of thermal energy of not more than 1500.degree. C. as shown in FIG. 128B to obtain a single-crystalline Si ingot 3502, which in turn is cut into a piece 3503 as shown in FIG. 128C and ground as shown in FIG. 128D, to obtain an Si single-crystalline substrate (wafer) 3504 with addition of a p-type impurity, as shown in FIG. 128E. When an n-type Si semiconductor layer 3505 shown in FIG. 129B is formed on an upper surface of an Si single-crystalline substrate (wafer) 3504 shown in FIG. 129A to form a p-n junction, on the other hand, employed is vapor phase diffusion employing PoCl.sub.3, application diffusion employing TiO.sub.2, SiO.sub.2 or P.sub.2 O.sub.5, or ion implantation of directly doping the wafer with P.sup.+ ions. Then, a back electrode 3506 and front electrodes 3507 are formed as shown in FIG. 129C, to complete a single-crystalline Si solar cell. However, this method has such a problem that the cost is increased due to employment of large amounts of power and Si materials. Under the present circumstances, the point of the study resides in how to cut the substrate in a small thickness in cutting of the ingot 3502 shown in FIG. 128C and how to reduce loss (cutting allowance) of Si in such slicing, in addition to automation and continuation.
In a single-crystalline Si solar cell, conversion efficiency is improved to cover the high cost of a bulk silicon substrate. In order to prevent reduction of efficiency caused by carrier recombination in the vicinity of a back surface of a substrate, a BSF (back surface field) type solar cell is prepared by introducing an internal electric field into a back surface of a substrate 3504 as shown in FIG. 130A. A p.sup.+ layer 3508 is formed on the back surface of the p-type substrate 3504 to provide an internal electric field for accelerating carriers which are generated in the vicinity of the back surface by this electric field so that the same can be effectively extracted as power. Due to this structure, photosensitivity is increased particularly with respect to longer wavelengths, to improve conversion efficiency to about 15%.
FIG. 130B shows a violet cell, which is adapted to prevent reduction of efficiency caused by carrier recombination on a front side (photoreceiving side) of a solar cell. The thickness of an n.sup.+ layer 3505 of the solar cell, which is 0.3 to 0.5 .mu.m in general, is reduced by etching to 0.1 to 0.2 .mu.m, thereby preventing carrier recombination in the vicinity of the n.sup.+ layer 3505 and improving conversion efficiency.
FIG. 130C shows a CNR solar cell (Comsat non-reflective solar cell) for preventing reduction of efficiency caused by reflection of light on a surface of the solar cell. According to this solar cell, large amounts of fine pyramidal irregularities of 1 to 2 .mu.m are formed on surfaces of a substrate 3504 and an n.sup.+ layer 3505, to reduce surface reflection by multiple reflection. Due to this structure, conversion efficiency of 18%, which is at the maximum level in relation to an Si solar cell, is obtained. The thickness of the n.sup.+ layer 3505 is set at 0.2 to 0.3 .mu.m here.
FIG. 131 shows relation between a light passage distance and a ratio of light to incident light as to single-crystalline silicon which receives and absorbs sunlight (AM-O) in the exoatmosphere on the average revolution orbit of the earth. This relation shows that single-crystalline silicon of 30 .mu.m absorbs 80% of incident light, and 100 .mu.m is required for absorbing 90% of the incident light with the remaining 10%. Since contribution to absorption is thus reduced with separation from the surface, there is an idea of employing a silicon plate which is thin to some extent. FIG. 132 shows relation between substrate thicknesses of various solar cells and conversion efficiency levels as reached. Referring to FIG. 132, marks + show optimum points which can sufficiently satisfy maximum values in employment of the respective materials. This figure shows that an extremely small thickness of not more than 1 .mu.m is sufficient for amorphous Si (a-Si) having a large absorption coefficient. When a bulk single-crystalline Si substrate is employed, its conversion efficiency cannot exceed that of amorphous Si unless its thickness is in excess of 10 .mu.m, and a thickness of about 70 to 80 .mu.m is required at the optimum point. A solar cell formed with such a thickness can attain conversion efficiency which is close to 20%. Even if the thickness is increased in ekcess of 80 .mu.m, however, further improvement of the conversion efficiency cannot be expected in comparison with a single-crystalline Si substrate of about 70 to 80 .mu.m in thickness. In general, however, a single-crystalline substrate requires a wafer thickness of about 250 .mu.m and cutting allowance of about 150 .mu.m, and 25 wafers can be generally cut out from an ingot of 1 cm in length. While various cutting methods are studied, a total thickness of 300 .mu.m including a wafer thickness of 200 .mu.m and cutting allowance of 100 .mu.m is the current limit. Namely, it is impossible to produce the aforementioned wafer of 70 to 80 .mu.m in thickness, and it is necessary to employ a wafer of 200 to 400 .mu.m in thickness in practice. Thus, it is difficult to reduce the cost, due to loss of the material.
As to a silicon substrate which is applied to a solar cell, consideration is made on a way to provide a material at the minimum cost by preparing the material from low-purity substandard portions caused on upper and lower end portions of an ingot in a process of manufacturing a wafer for an LSI. However, substandard portions which are obtained from LSI wafer production with dependence of the substrate material on byproducts for LSI use are about 10%, and this leads to a problem of insufficiency in material if only the substandard portions are employed.
2) Polycrystalline Si Solar Cell
A single-crystalline Si solar cell is disadvantageous in such points that the manufacturing process therefor is complicated, manufacturing energy is at a high level and a large amount of Si materials are employed. In order to solve these problems, there have been developed cast methods shown in FIGS. 133 and 134 (Prior Arts 5 and 7), and a ribbon method shown in FIG. 135 (Prior Art 6) for manufacturing polycrystalline Si solar cells. As shown in FIG. 133, a general cast method (Prior Art 5) is adapted to cool (slowly cool) and solidify an Si solution 3511 in a crucible 3512. This method can attain higher productivity as compared with single-crystalline Si, while the as-obtained conversion efficiency is at a relatively high level of 10 to 16%, although this level is slightly inferior to that of single-crystalline Si. FIG. 134 shows a newly proposed cast method (Prior Art 7), which has been developed by New Energy and Industrial Technology Development Organization (NEDO) as a part of Japan Sun Shine Project. An Si solution 3511 which is contained in a crucible 3512 is slowly cooled with the crucible 3512 from above, to obtain a polycrystalline Si ingot of the same shape as the crucible 3512. Referring to FIG. 134, numeral 3531 denotes a heater, numeral 3514 denotes a heat shielding plate, numeral 3515 denotes a support shaft, and numeral 3516 denotes a chamber. A polycrystalline Si solar cell can be formed by slicing the ingot manufactured through the aforementioned step, through a step similar to that for single-crystalline Si.
The ribbon method is adapted to directly obtain a polycrystalline sheet 3521 which is required for a solar cell from an Si solution, as shown in FIG. 135. The raw material can be effectively utilized since there is no need to slice an ingot, dissimilarly to single crystal or cast method polycrystalline Si. In principle, ribbon polycrystalline Si is lifted through a capillary die 3523 which is uprightly provided in a crucible 3522 containing the Si solution. The as-lifted ribbon polycrystalline Si is cut into a proper size with a laser beam, to be worked into a solar cell through a step similar to that for single-crystalline Si. Conversion efficiency is 7 to 14%, which is slightly slower as compared with that obtained by the cast method.
The polycrystalline Si solar cell also has a problem in cost in a point that a large amount of Si raw material must be employed, similarly to the single-crystalline Si solar cell. A general technical problem in Prior Art 6 resides in "how to cut a substrate in a small thickness and how to reduce loss in slicing", since the cost for a substrate material occupies about 25% of the cost for the overall solar cell in a high ratio.
In the cast method, the ingot is sliced with a multi-wire saw, to cause loss of about 120 to 150 .mu.m with respect to a substrate of 400 .mu.m. Namely, the Si raw material is used by 520 to 550 .mu.m, and the future technical subject resides in suppression of the same up to 300 .mu.m. Even if such reduction in thickness is implemented, however, the raw material is consumed about 10 to 100 times as compared with an amorphous Si solar cell as described below or the like.
3) Amorphous Si Solar Cell
An amorphous Si solar cell has excellent characteristics in cost, resource, manufacturing energy and environmental pollution, although the same is still disadvantageous in conversion efficiency and reliability.
The amorphous Si solar cell is manufactured by a method which is absolutely different from that for a solar cell employing a crystalline bulk Si substrate. FIGS. 136 and 137 schematically show a single-chamber reaction furnace method (Prior Art 8) and a continuous separation plasma reaction method (Prior Art 9) respectively. In either method, a raw material gas 3526 such as monosilane (SiH.sub.4) is introduced into a reaction chamber 3524 which is maintained in a vacuum of 0.1 to 10 Torr, and an RF electric field 3525 is applied to decompose the raw material gas 3526, thereby forming an amorphous Si layer 3528 on a substrate 3527 of glass or stainless steel. At this time, p-type amorphous Si or n-type amorphous Si is formed when B.sub.2 H.sub.6 (diborane) 3529 or phosphine (PH.sub.3) 3530 is mixed into the raw material gas 3526. Referring to FIG. 137, numeral 3531 denotes a shutter. Thus, a p-n junction can be formed by simply switching the raw material gas 3526. The amorphous Si solar cell requires small manufacturing energy, and the Si layer forming an element is reduced to 1/100 as compared with single-crystalline and polycrystalline bulk silicon solar cells, to implement reduction in cost. Further, the film formation temperature is so reduced that a low-priced substrate of glass or stainless steel can be employed, to also effectively reduce the cost.
The amorphous Si solar cell is different in principle of power generation and structure from a crystalline Si solar cell. In general, crystalline silicon has a p-n structure as shown in FIG. 138, and photocarriers are moved by diffusion. On the other hand, the amorphous Si solar cell has a p-i-n (p-type - intrinsic - n-type) structure which is prepared by interposing an intrinsic semiconductor layer (i) between a p-type layer (p) and an n-type layer (n) as shown in FIG. 139, and photocarriers are mainly moved by drifting by an electric field contained in the solar cell. Referring to FIG. 139, TCO denotes a transparent conductive oxide film which serves as an electrode, and Gs denotes a glass plate. Due to such provision of the intrinsic semiconductor layer (i), most of electrons and holes are generated by light in a portion having the internal electric field, i.e., a depletion layer, since a space-charge region (depletion layer) in a boundary of an ordinary p-n junction is wide and hence the electrons and holes can be immediately extracted as a photogenerated current. Namely, it is possible to omit a process of movement of minority carriers to a p-n boundary region by diffusion in an ordinary p-n junction. In this case, however, part of incident light is absorbed by a p-type doped layer before reaching the intrinsic semiconductor layer (i) and not effectively utilized for power generation since amorphous Si has a large absorption coefficient while photocarriers are generated mainly in the intrinsic semiconductor layer (i) in the amorphous Si solar cell. FIG. 140 shows comparison of light absorption coefficients with respect to wavelengths (energy of photons) of light in relation to amorphous Si and crystal silicon etc., and FIG. 141 shows comparison of light absorption spectra of amorphous Si:H mixed with hydrogen and single-crystalline Si. It is understood from FIGS. 140 and 141 that the amorphous Si or the amorphous Si:H has a larger absorption coefficient by about 1 digit as compared with the single-crystalline Si in the vicinity of 0.5 .mu.m forming the peak of solar energy. It is also understood that the single-crystalline Si has a larger coefficient absorption by one digit than the amorphous Si on a longer wavelength side beyond the wavelength of 0.7 .mu.m.
FIG. 142 shows light absorption states with parameters of light absorption coefficients, by plotting rates of absorption of photons with respect to distances of light passage while setting values of absorption coefficients. FIG. 142 shows four curves with light absorption coefficients of 10.sup.2, 10.sup.3, 10.sup.4 and 10.sup.5. It is understood that the amorphous Si can be reduced in thickness as compared with the single-crystalline silicon and is suitable for a thin film structure since a large quantity of light is absorbed in a short distance when the light absorption coefficient is at a high level.
The above description is summarized as follows:
1. An extremely thin layer is sufficient for an amorphous Si type, which has a large absorption coefficient. However, its conversion efficiency is maximum at 14% in theory as shown in FIG. 132, and is extremely low as compared with the conversion efficiency of single-crystalline Si exceeding 20%.
2. While photocarriers are mainly generated in the intrinsic semiconductor layer (i) shown in FIG. 139 in the amorphous Si type, part of incident light is absorbed by the p-type doped layer before reaching the intrinsic semiconductor layer (i), and is not effectively used for power generation. It is important to polycrystallize or single-crystallize the doped layer in order to reduce the doped layer on the incident light side and increase the quantity of light reaching the intrinsic semiconductor layer (i), while it is difficult to attain this. While it is necessary to reduce a process temperature particularly when a low-priced glass substrate or the like is employed, it is difficult to implement this by a conventional method since Si has a high crystallization temperature.
3. In a solar cell employing amorphous Si, remarkable improvement of conversion efficiency can be expected if it is possible to arrange a cell having sensitivity to shorter wavelengths on a light input side (front side) and to arrange a cell having sensitivity to longer wavelengths on a back side, since the wide wavelength range of the sunlight can be effectively utilized. To this end, conversion efficiency is improved if it is possible to arrange amorphous Si on a front cell and to arrange a single-crystalline or polycrystalline silicon thin film on a back side. In order to implement such a structure in the prior art, however, it is necessary to employ a bulk silicon single-crystalline substrate. Thus, a low-priced glass substrate or the like cannot be employed and the substrate is increased in thickness, to sacrifice cost reduction.
4. In a cell having excellent absorption with a large absorption coefficient, it is necessary to effectively extract minority carriers among electrons and holes generated by light in the vicinity of the surface, i.e., electrons in a p-type semiconductor or holes in an n-type semiconductor. To this end, it is important to reduce crystal defects or to reduce the depth of a p-type layer so that the as-generated minority carriers can be effectively moved to a depletion layer to be extracted, since a large amount of crystalline defects are inevitably caused in a doped layer (p-type layer in a p-i-n structure) having high concentration of an impurity in the surface part such that electrons and holes are immediately recombined and annihilated. Although a general amorphous Si film contains hydrogen which is coupled with uncoupled ones to reduce defects, defect density is higher by 3 to 5 digits than single-crystalline Si to cause significant loss by surface recombination. Amorphous Si has defect density of 10.sup.20 /cm.sup.3, amorphous Si (amorphous Si:H) hydrogenated by glow discharge (plasma CVD) has defect density of 10.sup.15 to 10.sup.17 /cm.sup.3, and single-crystalline Si has defect density of 10.sup.12 /cm.sup.3.
FIG. 143 shows physical constants of amorphous Si:H and single-crystalline Si. As understood from FIG. 143, electric characteristics of the amorphous Si:H such as mobility of electrons and holes, diffusion length, life times of electrons and holes and the like are inferior to those of single-crystalline silicon, leading to inferior conversion efficiency.
Thus, it is preferable to arrange a crystalline silicon thin film layer on a front side, to form an intrinsic semiconductor layer by an amorphous Si thin film and to arrange a crystalline silicon thin film having an excellent absorption coefficient for a longer wavelength side on a back side, in order to manufacture a solar cell having a smaller amount of silicon material (reduction in thickness) and excellent conversion efficiency by effectively utilizing optical and electric characteristics. In general, however, it is difficult to form a single-crystalline or polycrystalline silicon thin film on an amorphous Si thin film since a single-crystalline or polycrystalline thin film cannot be formed on a low-priced glass substrate and single-crystalline Si can be formed only on a single-crystalline substrate in general.
FIGS. 144 to 147 show exemplary proposals (Prior Arts 10 to 13) in relation to a solar cell of amorphous silicon carbide (amorphous SiC) having sensitivity to short-wave light, which is arranged on a front side on a base of an amorphous Si layer or the like. Referring to each of FIGS. 144 to 147, numeral 3535 denotes a glass substrate, numeral 3536 denotes a transparent conductive oxide (TCO) film, numeral 3537 denotes a p-type amorphous SiC film, numeral 3538 denotes an intrinsic amorphous Si layer, numeral 3539 denotes an n-type amorphous Si film, and numeral 3541 denotes a back-side electrode of aluminum or silver. Referring to FIG. 145, numeral 3542 denotes an n-type microcrystal Si layer, numeral 3543 denotes a TiO.sub.2 layer and numeral 3544 denotes a semi-texture structure (SuS) layer, while numeral 3545 in FIG. 146 denotes a graded hand-gap amorphous SiC layer, and numeral 3546 in FIG. 147 denotes a highly qualified intrinsic amorphous Si layer. As shown in FIG. 148, performed is improvement/research of arranging an intrinsic amorphous SiGe layer 3547 on a back side (Prior Art 14). However, the absorption coefficient is lowered in relation to longer wavelengths as shown in FIG. 141 since all these layers are prepared from amorphous materials. Thus, this technique is not yet put into practice due to problems of electric characteristics in relation to covering of a wide frequency band. While a polycrystalline Si layer 3548 is arranged on a back end in Prior Art 15 as shown in FIG. 149, the material for the substrate is restricted to a high-priced one in this case since the substrate material must withstand high-temperature environment for crystallizing the polycrystalline Si film 3548.
(Problem of Prior Arts 4 to 15)
The aforementioned problems of Prior Arts 4 to 15 are summarized as follows:
1) A crystalline Si solar cell (Prior Arts 4 to 7) requires a substrate thickness of 200 .mu.m at the minimum and 300 .mu.m in general, since a bulk silicon wafer must be employed. Further, the silicon wafer is cut out from an ingot and hence cutting loss of 120 to 150 .mu.m is caused leading to requirement for an Si material of 320 to 450 .mu.m in thickness as a whole.
In order to attain conversion efficiency of 20% which is close to the theoretical value, loss of the material is caused since the crystalline Si film may have a thickness of 70 to 80 .mu.m, leading to increase in cost.
It is necessary to reduce the thickness of the silicon wafer to minimize loss in cutting in order to reduce the cost, while it is difficult to reduce the thickness below 200 .mu.m in view of the yield, and it is also difficult to reduce the loss in cutting below 100 .mu.m, so far as a multi-wire saw or the like is employed. Thus, cost reduction for the solar cell is limited.
2) In the crystalline Si substrate (Prior Arts 4 to 7), which is mainly prepared by a method of utilizing substandard portions on upper and lower ends of an ingot for an LSI for reducing the cost, the resource therefor tends to be insufficient in general.
3) In the amorphous Si solar cell (Prior Arts 8 to 15), cost reduction can be implemented since the light absorption coefficient in the vicinity of the peak (500 .ANG.) of solar energy is higher by at least one digit as compared with a single-crystalline Si solar cell and hence a larger quantity of light is absorbed in a short distance and the film thickness can be reduced to not more than 1/100 as compared with the single-crystalline solar cell. Further, the amorphous Si is effective for cost reduction since the film forming temperature is at a low level and a low-priced glass substrate can be employed. However, the same has a large amount of crystal defects which is larger by 3 to 5 digits as compared with the single-crystalline Si solar cell, while electrons and holes are easily annihilated by recombination, mobility is lower by 4 to 5 digits leading to inferior conversion efficiency, and the same is easily deteriorated by time change.
(Prior Art 16)
Conventional semiconductor devices (semiconductor memory devices) for serving as mass storage devices include a volatile dynamic random access memory (hereinafter referred to as a DRAM) and an electrically writable/erasable nonvolatile memory such as a flash EPROM.
FIG. 207 is a sectional view showing a semiconductor device for serving as a DRAM according to Prior Art 16, and FIG. 208 illustrates an internal circuit of a ferroelectric memory device employing ferroelectric substances for capacitor parts by Ramtron Corp., for example. Referring to FIGS. 207 and 208, numeral 3601 denotes a p-type semiconductor substrate, numeral 3602 denotes a transistor part, numeral 3603 denotes a bit line, numeral 3604 denotes a LOCOS oxide film, numeral 3605 denotes a polycrystalline (polysilicon/polycide) word line, numeral 3606 denotes a first insulating film, numeral 3607 denotes a capacitance part (data holding part), numeral 3608 denotes a first wire connecting the transistor part 3602 with the capacitance part 3607, numeral 3609 denotes a dielectric film, numeral 3610 denotes an upper electrode of the capacitance part 3607, and numeral 3611 denotes a plate.
In the conventional memory device employing ferroelectric substances for capacitors, a 1-bit memory cell is formed by two transistors and two capacitors as shown in FIG. 208 due to remarkable dispersion in charge quantity caused by the dielectric constant and polarization of the capacitance part 3607, and the capacitors are polarized in opposite directions to guarantee data by detecting the difference therebetween.
In general, the dielectric film 3609 of the capacitance part 3607 is formed by a ferroelectric film. In such a ferroelectric film, dielectric constants of crystals are increased with temperature reduction to cause phase transition by divergence at a certain critical temperature (Curie temperature) thereby causing spontaneous dielectric polarization in a low-temperature phase. Such a ferroelectric film is prepared from PZT consisting of mixed crystals of PbZrO.sub.3 and PbTiO.sub.3 or BST consisting of mixed crystals of BaTiO.sub.3 and SrTiO.sub.3. While such a material is in a perovskite or pyrochlore structure, it is necessary to utilize a perovskite crystal system since a pyrochlore crystal system has a low dielectric constant. In general, the aforementioned ferroelectric film is generally formed by a polycrystalline thin film.
(Prior Art 17)
FIG. 209 is a sectional view showing a semiconductor device for serving as a nonvolatile memory device (EPROM/flash EPROM) according to Prior Art 17. Referring to FIG. 209, numeral 3621 denotes a p-type semiconductor substrate, numeral 3622 denotes a LOCOS oxide layer (SiO.sub.2), numeral 3623 denotes a gate insulating film, numeral 3624 denotes a floating gate, numeral 3625 denotes a control gate, and numeral 3626 denotes an interlayer insulating film which is interposed between the gates 3624 and 3625. The floating gate 3624 is prepared from polycrystalline silicon (polysilicon). The interlayer insulating film 3626 is formed by a multilayer film (oxi-nitride: ONO) consisting of silicon oxide (SiO.sub.2) and silicon nitride (SiN) or independently prepared from silicon oxide (SiO.sub.2), so that an SiO.sub.2 layer is in contact with the floating gate 3624 in either case. In formation of this interlayer insulating film 3626, the polysilicon forming the floating gate 3624 is oxidized to grow SiO.sub.2.
(Prior Art 18)
FIG. 210 is a sectional view showing a semiconductor device for serving as a nonvolatile memory device (EPROM/flash EPROM) according to Prior Art 18. Referring to FIG. 210, members having the same functions as those in Prior Art 17 are denoted by similar reference numerals. In the semiconductor device according to Prior Art 18 which aims at a higher degree of integration as compared with that according to Prior Art 17, a source 3627 and a drain 3628 are formed by embedding/diffusion to form a transistor. Also in the semiconductor device according to Prior Art 18, a floating gate 3624 is prepared from polycrystalline silicon (polysilicon) and an interlayer insulating film 3626 is formed by a multilayer film (ONO) consisting of silicon oxide (SiO.sub.2) and silicon nitride (SiN) or independently prepared from silicon oxide (SiO.sub.2), so that an SiO.sub.2 layer is in contact with the floating gate 3624 in either case. In formation of this interlayer insulating film 3626, the polysilicon forming the floating gate 3624 is oxidized to grow SiO.sub.2, similarly to Prior Art 17.
(Prior Art 19)
FIG. 212 is a circuit diagram of a memory cell forming a memory array of a general DRAM according to Prior Art 19. Referring to FIG. 212, symbol 3632 denotes a transistor part, symbol 3633 denotes a bit line, symbol 3635 denotes a word line, and symbol 3637 denotes a capacitance part (data holding part). According to Prior Art 19, a 1-bit memory cell can be formed by a single transistor part and a single capacitance part, whereby it is possible to attain relatively large capacitance.
In such a DRAM which is a volatile memory device, however, information as stored is lost when power is removed. Therefore, it is necessary to regularly supply power in order to hold the stored information, and the application range of this DRAM is disadvantageously limited.
(Prior Art 20)
FIG. 213 is a sectional view showing a memory cell (memory transistor) of a general flash EPROM according to Prior Art 20. Referring to FIG. 213, symbol 3641 denotes a semiconductor substrate, symbol 3642 denotes a floating gate which is formed above the semiconductor substrate 3641, symbol 3643 denotes a control gate which is formed above the floating gate 3642, and symbols 3644 and 3645 denote a source diffusion region and a drain diffusion region which are selectively formed on an upper surface of the semiconductor substrate 3641. A thin gate insulating film (oxide film) is formed in a clearance 3646 between the floating gate 3642 and the semiconductor substrate 3641 for moving electrons through a tunnel phenomenon.
Operation of the flash EPROM is now described. First, the source diffusion region 3644 is grounded and a program voltage is applied to the drain diffusion region 3645 while a voltage is applied to the control gate 3643, whereby the memory cell enters an ON state to allow a current flow. At this time, avalanche breakdown takes place in the vicinity of the drain diffusion region 3645, to generate electron-hole pairs. The holes flow to the ground potential through the semiconductor substrate 3641, while the electrons flow toward a channel 3629 direction into the source diffusion region 3644. However, parts of the electrons are accelerated by an electric field which is developed across the floating gate 3642 and the drain diffusion region 3645, to be injected into the floating gate 3642. Consequently, the threshold voltage of the memory cell is increased. In this state, information "0" is stored. In order to erase the information, on the other hand, the drain diffusion region 3645 is opened, the control gate 3643 is grounded and a voltage is applied to the source diffusion region 3644. Then, a tunnel phenomenon is caused by an electric field which is developed across the source diffusion region 3644 and the floating gate 3642, to extract the electrons from the floating gate 3642. Consequently, the threshold voltage of the memory cell is reduced. In this state, information "1" is stored.
Such a flash EPROM can be regarded as having a higher possibility for mass storage than a DRAM since a 1-bit memory cell can be formed by a single memory transistor. However, this device has a problem of a slow data writing/erasing time. Further, a high electric field is applied to the memory transistor in data writing/erasing to inject or extract electrons through the gate insulating film which is formed in the clearance 3646. Thus, the gate insulating film provided in the clearance 3646 is fatigued with data writing and erasing operations to be gradually reduced in remanence and finally broken. Thus, the number of writing/erasing times is limited.
(Prior Art 21)
In order to solve the aforementioned problems of Prior Art 16 and Prior Art 20, Ramtron Corp. has developed a nonvolatile DRAM (hereinafter referred to as an FRAM) in which a dielectric film of a capacitance part is made of a ferroelectric substance (refer to Nikkei Microdevices, June 1992, pp. 78 to 83 and Japanese Patent Laying-Open Gazettes Nos. 64-66897 (1989), 64-66899 (1989), 1-278063 (1989) and 2-113496 (1990)).
FIG. 214 is a sectional view showing an exemplary ferroelectric DRAM according to Prior Art 21, in which a capacitor part (capacitance part) is made of a ferroelectric substance. An internal circuit of this semiconductor memory device is similar to that of Prior Art 16 shown in FIG. 208, and members having the same functions as those shown in FIG. 208 are denoted by the same reference numerals in the following description. Referring to FIGS. 208 and 214, numeral 3601 denotes a p-type semiconductor substrate, numeral 3602 denotes a transistor part, numeral 3603 denotes a bit line, numeral 3604 denotes a LOCOS oxide film, numeral 3605 denotes a polycrystalline (polysilicon/polycide) word line, numeral 3606 denotes a first insulating film, numeral 3607 denotes a capacitance part (data holding part), numeral 3608 denotes a first wire connecting the transistor part 3602 with the capacitance part 3607, numeral 3609 denotes a dielectric film which is made of a ferroelectric substance, numeral 3610 denotes an upper electrode of the capacitance part 3607, and numeral 3611 denotes a plate.
In the memory device according to Prior Art 21 employing a ferroelectric substance for a capacitor, a 1-bit memory cell is formed by two transistors and two capacitors as shown in FIG. 208 due to remarkable dispersion in charge quantity caused by the dielectric constant and polarization of the capacitance part 3607, and the capacitors are polarized in opposite directions to guarantee data by detecting the difference therebetween.
In such a ferroelectric film which is employed for forming the dielectric film 3609 of the capacitance part 3607, dielectric constants of crystals are increased with temperature reduction to cause phase transition by divergence at a certain critical temperature (Curie temperature) thereby causing spontaneous dielectric polarization in a low-temperature phase. Such a ferroelectric film is prepared from PZT consisting of mixed crystals of PbZrO.sub.3 and PbTiO.sub.3 or BST consisting of mixed crystals of BaTiO.sub.3 and SrTiO.sub.3. While such a material is in a perovskite or pyrochlore structure, it is necessary to utilize a perovskite crystal system since a pyrochlore crystal system has a low dielectric constant. In order to implement such a ferroelectric substance having a perovskite crystal structure, an electrode of the ferroelectric substance is generally in a multilayer film structure in which a Pt layer is arranged in an interface with the ferroelectric substance, so that no electrode material (particularly a metal material) enters crystals of the ferroelectric substance to cause a leakage current or deterioration of film characteristics, thereby attaining excellent crystallinity of the ferroelectric substance. In Prior Art 21, the aforementioned ferroelectric film is generally prepared from an amorphous film.
(Prior Art 22)
FIG. 215 illustrates a semiconductor device for serving as a nonvolatile memory device (EPROM/flash EPROM) according to Prior Art 22. As shown in FIG. 215, the semiconductor device according to Prior Art 21 is a ferroelectric gate field effect transistor (hereinafter referred to as MFSFET) in which a memory cell is formed by a transistor 3652 having a gate insulating film 3651 of a ferroelectric substance (refer to Japanese Patent Laying-Open Gazette No. 4-192173 (1992), for example). Referring to FIG. 215, symbol 3653 denotes a p-type semiconductor substrate, symbol 3654 denotes a gate electrode, symbol 3655 denotes a source and symbol 3656 denotes a drain.
Also in the semiconductor device according to Prior Art 22, the ferroelectric film is prepared from PZT consisting of mixed crystals of PbZrO.sub.3 and PbTiO.sub.3 or BST consisting of mixed crystals of BaTiO.sub.3 and SrTiO.sub.3. Further, the ferroelectric film is generally prepared from an amorphous film also in the semiconductor device according to Prior Art 22.
(Problem of Prior Art 16)
When the ferroelectric film 3609 for the capacitance part 3607 is formed by a polycrystalline or amorphous thin film in Prior Art 16, a polarization field curve (PE hysteresis curve) with respect to an external electric field is rhomboidally dulled as shown in FIG. 211, to cause dispersion such that remanence is reduced and the dielectric constant is also reduced.
Further, a metal forming the electrode may enter between crystal grains of the ferroelectric film 3609, to cause a leakage current or deterioration in durability. Although there has been proposed a method of depositing a thin RuO.sub.2 film on the electrode or adding about several to 10% of La to PZT in order to solve this problem, the dielectric constant is reduced by about 1 digit in either case. While PZT is held by multilayer metals including a Pt layer to be improved in crystallinity in the ferroelectric memory device by Ramtron Corp., for example, the fabrication cost is increased in this case. Thus, none of the aforementioned methods is sufficiently practicable.
As a material for the dielectric film 3609 for the capacitance part 3607 of the conventional DRAM, SiO.sub.2, Si--N, Ta.sub.2 O.sub.5 or the like has been put into practice or studied. The capacitance part generally requires storage capacitance of about 20 to 40 fF, which remains substantially unchanged upon refinement of the device. In a mass storage DRAM, therefore, a capacitive surface area is increased by a complicated structure such as a stack, trench or fin type structure to suppress the cell size. When such a complicated capacitance structure is implemented, however, the number of fabricating steps is inevitably increased. In recent years, therefore, study has been developed to apply a polycrystalline dielectric film having a high dielectric constant to the dielectric film 3609 of the capacitance part 3607. In this case, it is necessary to regulate the crystalline structure in order to suppress dispersion in dielectric constant thereby stabilizing characteristics, while such stabilization of characteristics is limited in a polycrystalline or amorphous thin film. A film thickness of at least 2000 .ANG. is required in order to obtain a stable film, while the area of the capacitance part 3607 must also be increased in this case and hence it is impossible to cope with refinement of the device.
(Problems of Prior Art 17 and Prior Art 18)
For example, a flash EPROM has a high data reading speed of about several 10 to 200 nsec. in general, while a data writing or erasing operation requires an extremely long time of several .mu.sec. to several msec. depending on the system, and there has been strong requirement for increase in speed. In the semiconductor device having a two-layer structure provided with a floating gate according to each of Prior Art 17 and Prior Art 18, it is necessary to increase a coupling coefficient (GCR) in order to improve writing efficiency etc.
Assuming that C.sub.1 represents capacitance between the floating gate 3624 and the semiconductor substrate 3621 and C.sub.2 represents that between the floating gate 3624 and the control gate 3625, the coupling coefficient (GCR) is in proportion to C.sub.2 /(C.sub.1 +C.sub.2). Namely, it is effective to maximize the capacitance C.sub.2 between the floating gate 3624 and the control gate 3625. This can be attained by the following three methods:
(1) To reduce the thickness of the interlayer insulating film 3626 which is provided between the gates 3624 and 3625.
(2) To increase the dielectric constant of the interlayer insulating film 3626.
(3) To increase the overlap area of the gates 3624 and 3625.
In practice, however, the device has the following restrictions:
[1] The thickness of the interlayer insulating film 3626 is decided due to requirement in performance for ensuring withstand voltages of about 30 to 40 V for the floating gate 3624 and the control gate 3625. Namely, the thickness must be maintained at a certain degree since the interlayer withstand voltage is disadvantageously reduced if the thickness is reduced. In more concrete terms, the gate insulating film 3623 has a thickness (distance between the floating gate 3624 and the semiconductor substrate 21) of about 20 nm and the interlayer insulating film 3626 has a thickness (distance between the floating gate 3624 and the control gate 25) of about 60 nm (SiO.sub.2 : 10 nm and SiN: 50 nm in ONO structure) in a general EPROM.
[2] The material for the interlayer insulating film 3626 is restricted to an ONO or SiO.sub.2 film, and hence it is impossible to increase the dielectric constant of the interlayer insulating film 3626 so far as such a material is employed. In more concrete terms, SiO.sub.2 and ONO films have dielectric constants of about 3.9 and 6 to 9 respectively. It may conceivably be possible to employ a film of BST consisting of mixed crystals of BaTiO.sub.3 and SrTiO.sub.3) or PLZT obtained by adding La to mixed crystals of PbZrO.sub.3 and PbTiO.sub.3 as a high dielectric film. This material has a dielectric constant of about 300, which is higher by 30 to 50 times than that of the ONO film. In a conventional fabrication method, however, a thickness of at least 200 nm (300 to 400 nm in practice) is required in order to stabilize the properties of the film, and hence it is necessary to hold the film with multilayer metals including a Pt layer or the like in order to improve crystallinity of the film. Thus, this film is unsuitable for the interlayer insulating film 3626 provided between the gates 3624 and 3625 of the memory device, and is not put into practice.
[3] In each of Prior Art 17 and Prior Art 18, the area S1 of the floating gate 3624 is substantially doubled as compared with the area S2 of the channel region so that the capacitance C.sub.2 between the floating gate 3624 and the control gate 3625 is larger than the capacitance C.sub.1, in order to attain a coupling coefficient of about 0.6. While the capacitance C.sub.2 is increased when the overlap area between the floating gate 3624 and the control gate 3625 is thus increased, however, the memory cell area is inevitably increased contrarily to the requirement for miniaturization of the device. In order to attain a coupling coefficient (GCR) of 0.6 to 0.7, the overlap area between the floating gate 3624 and the control gate 3625 is doubled or tripled as compared with the area of the channel region. While it is necessary to reduce the overlap area between the floating gate 3624 and the control gate 3625 in order to reduce the memory cell area, the coupling coefficient cannot be reduced below 0.5, in order to maintain performance of the memory device. In particular, it is possible to reduce the cell area by separating adjacent cells from each other by a trench T (see FIG. 172). In this case, however, the area S1 of the floating gate 3624 must be substantially equalized to the area S2 of the channel region in design, and the coupling coefficient is deteriorated to about 0.4 in this case. Particularly when the semiconductor device according to Prior Art 18 is formed not by embedding/diffusion but N.sup.+ injection with a mask of the floating gate 3624 in self alignment, the size of the floating gate 3624 is inevitably equalized to that of the channel region. In this case, the coupling coefficient is extremely reduced.
In each of Prior Art 17 and Prior Art 18, further, the floating gate 3624 is made of polysilicon and grown by oxidation to form the interlayer insulating film 3626, and hence these layers are inferior in crystallinity and not dense in film quality.
Further, it is difficult to form the floating gate 3624 in a flat state due to the polycrystalline structure of its surface, and the floating gate 3624 is generally provided with projections. When the floating gate 3624 is oxidized in this state, the projections are further grown due to a high oxidation speed. These projections extremely reduce the withstand voltage of the interlayer insulating film 3626 provided between the floating gate 3624 and the control gate 3625. Therefore, it is necessary to smooth such projections before oxidation of the floating gate 3624 by implanting Ar.sup.+ or As.sup.+ into the polysilicon at about 50 keV and bringing the polysilicon surface of the floating gate 3624 into an amorphous state. Thus, the processing steps are increased to increase the fabrication time.
(Problem of Prior Art 19)
In the DRAM according to Prior Art 19 which is a volatile memory device, information as stored is lost when power is removed as hereinabove described. Therefore, it is necessary to regularly supply power in order to hold the stored information, and the application range of this DRAM is limited.
While a 1-bit memory cell is formed by a single transistor part and a single capacitance part in the semiconductor device according to Prior Art 19, a concrete system for forming a 1-bit memory cell by only a single transistor part is desired in order to further miniaturize the memory cell.
(Problem of Prior Art 20)
In the semiconductor device according to Prior Art 20, the data writing/erasing is at a slow speed and the gate insulating film provided in the clearance 3646 is fatigued with data writing and erasing operations to be finally broken, as hereinabove described. Thus, the number of writing/erasing times is limited.
(Problem of Prior Art 21)
In the semiconductor device according to Prior Art 21, a 1-bit memory cell is formed by two transistors and two capacitors as shown in FIG. 208, and hence miniaturization of the memory cell is limited. Thus, a concrete system for forming a 1-bit memory cell by only a single transistor part is desired, as hereinabove described.
Further, a multilayer film forming step is complicated in the semiconductor device according to Prior Art 21 due to the multilayer film structure including a Pt layer, and hence simplification of the steps in mass production is limited.
(Problem of Prior Art 22)
Although a 1-bit memory cell is formed by a single transistor in the MFSFET according to Prior Art 22, internal crystallinity is inferior due to an amorphous film which is applied to the gate insulating film 3651 of a ferroelectric substance in particular, and hence its film thickness must be increased to some extent. In order to ensure a constant breakdown voltage for the gate insulating film 3651, therefore, a single transistor memory cell is increased in area. Thus, it is impossible to implement a practical degree of integration when a plurality of cells are arranged in parallel with each other.
In order to implement the MFSFET according to Prior Art 22, it is necessary to form a perovskite crystalline film of a ferroelectric substance on an upper surface of a silicon film (Si), a silicon oxide film (SiO.sub.2) or a silicon nitride film (Si--N) as an underlayer. Under the present circumstances, however, there has been announced no effective method of crystallizing a perovskite film of a ferroelectric substance, and it is extremely difficult to single-crystallize the same in particular. Therefore, improvement of polarization field characteristics of the ferroelectric substance with respect to an external electric field is limited. Particularly in formation of a thin film, a problem is caused in stability of characteristics.
(Prior Art 23)
FIG. 231 illustrates a general 6-transistor CMOS-SRAM cell. The SRAM is formed by nMOS transistors N1 and N2, a CMOS flip-flop in which an input and an output of a CMOS inverter pair connected with single ones of pMOS transistors P1 and P2 respectively are cross-connected with each other, and nMOS transfer transistors N3 and N4. When information is "1", the transistors P1 and N2 conduct while the transistors P2 and N1 enter nonconducting states. When information is "0", on the other hand, the transistors P1 and N2 enter nonconducting states while the transistors P2 and N1 conduct. Consequently, no consumption current flows in this cell in a standby state except a leakage current component, whereby power consumption can be saved in the standby state.
FIG. 232 is a sectional view showing a semiconductor device according to prior art 23, in which the pMOS transistor P1 provided in the SRAM shown in FIG. 231 is formed by a TFT. Referring to FIG. 232, numeral 3701 denotes a substrate, numeral 3702 denotes a wire to be connected to a gate of the nMOS transistor N1, numeral 3703 denotes a first insulating film, numeral 3704 denotes a gate of the pMOS transistor P1, numeral 3705 denotes a wire to be connected to gates of the pMOS transistor P2 and the nMOS transistor N2, numeral 3706 denotes a second insulating film, numeral 3707 denotes a drain of the transistor P1, numeral 3708 denotes a source of the transistor P1, and numeral 3709 denotes a channel interposed between the drain 3707 and the source 3708.
In the prior art 23, the wires 3702 and 3705, the gate 3704, the drain 3707, the source 3708 and the channel 3709 are formed by polycrystalline (polysilicon) or amorphous silicon films.
(Prior Art 24)
FIG. 233 is a sectional view showing a semiconductor device according to prior art 24, in which a TFT is built into a display device of an active matrix liquid display. Referring to FIG. 233, numeral 3711 denotes a condensed capacitance part, numeral 3712 denotes a TFT, numeral 3713 denotes a glass substrate, numeral 3714 denotes a lower electrode of aluminum or the like for the condensed capacitance part 3711, numeral 3715 denotes an Al.sub.2 O.sub.3 film, numeral 3716 denotes a transparent upper electrode (ITO) for the condensed capacitance part 3711, numeral 3717 denotes a gate of MoTa or the like for the TFT 3712, numeral 3718 denotes an Si channel for the TFT 3712, numeral 3719 denotes a source of n.sup.+ -type Si or the like for the TFT 3712, numeral 3721 denotes a drain of n.sup.+ -type Si or the like for the TFr 3712, and numeral 3722 denotes a dielectric film (gate insulating film) of SiN.sub.x or SiO.sub.2. FIG. 234 is an enlarged sectional view showing the TFT 3712. Referring to FIG. 234, numeral 3723 denotes an SiN.sub.x film for serving as a gate insulating film, numeral 3724 denotes an amorphous Si film, and numeral 3725 denotes an electrode of Mo/Al or the like. Referring to FIG. 234, the TFT 3712 is generally formed in order of the gate 3717, the gate insulating films 3722 and 3723, the amorphous Si film 3724, the Si channel 3718, the source 3719 and the drain 3721, and the electrode 3725. The gate insulating films 3722 and 3723, the lower electrode 3714 for the condensed capacitance part 3711, the gate 3717 and the Si channel 3718 are in polycrystalline or amorphous structures.
(Problem of Prior Art 23)
While the drain 3707, the source 3708 and the channel 3709 of the TFT according to the prior art 23 are formed by polycrystalline (polysilicon) or amorphous silicon films, the ratio of a drain current in an ON state of the transistor to that in an OFF state is small in a TFT employing polysilicon, for example. When an OFF-state current is reduced in order to reduce a standby current of the SRAM in this case, the ON-state current is also reduced such that it takes time to charge the potential of the memory cell immediately after writing from Vcc - Vth to Vcc, and hence the access time is retarded. When the ON-state current is increased, on the other hand, the OFF-state current is also increased to disadvantageously increase the standby current. In a TFT employing amorphous silicon which is mainly applied to an active element for a liquid crystal display, on the other hand, the consumption current comes into question when the TFT is applied to a portable device or a color display, due to a large OFF-state current.
(Problem of Prior Art 24)
In the prior art 24, on the other hand, the gate insulating films 3722 and 3723 which are in amorphous structures are so inferior in crystallinity that these films are increased in thickness to 400 to 500 nm on the TFT side and to at least 200 nm in the dielectric film 3722 on the side of the condensed capacitance part 3711 respectively. Therefore, a large area is required per element for obtaining a sufficient condensed capacitance value, leading to hindrance to high definition. In other words, improvement in numerical aperture is limited.
In the prior art 24, further, the S1 channel 3718 is provided in a polycrystalline or amorphous structure, and hence electron mobility in the channel is reduced while resistance values of the lower electrode 3714 for the condensed capacitance part 3711 and the TFT gate 3717 are increased when the same are prepared from general Ta or the like. Therefore, increase in operating speed of the TFT for a driver is limited.
When the lower electrode 3714 for the condensed capacitance part 3711 and the gate 3717 of the TFT 3712 as well as the gate insulating films 3722 and 3723 provided on the upper surfaces thereof are reduced in thickness as shown in FIG. 233, further, it is necessary to form the lower electrode 3714 and the gate 3717 serving as underlayers in flat states. When the lower electrode 3714 and the gate 3717 are made of Al or an Al alloy for attaining low resistance, however, polycrystalline or amorphous structures provided on surfaces thereof may partially define nuclei for growing projections (hillocks) in heat treatment of about 400.degree. C. Thus, a leakage current may be generated in employment. In the prior art 24, therefore, the surfaces of the lower electrode 3714 and the gate 3717 are covered with thick Al.sub.2 O.sub.3 layers by anodization as shown in FIG. 233, for preventing occurrence of hillocks. In this case, however, the number of members is increased with increase of the fabrication steps, and hence the cost is increased.
(Prior Art 25)
A liquid crystal display (LCD) comprising an electrically insulating transparent substrate which is provided thereon with pixel electrodes of liquid crystal elements and thin-film type active elements for driving the liquid crystal elements for respective pixels which are arranged in the form of a matrix is called an active matrix LCD, and forms the mainstream of commercially available liquid crystal displays. The active elements for driving the liquid crystal elements are typically formed by thin-film transistors (TFT).
FIG. 267 is a circuit diagram showing a circuit structure for one pixel in such an active matrix LCD according to prior art 25. Referring to FIG. 267, a liquid crystal element 3853 for one pixel and a TFT 3854 which is connected in series with this liquid crystal element 3853 for driving the same are arranged at an intersection between a signal line 3851 and a scanning line 3852 which are arranged in the form of a matrix. In this example, the TFT 3854 is formed by a MOS field-effect transistor (MOSFET). The series circuit of the liquid crystal element 3853 and the TFT 3854 is interposed between the signal line 3851 and the ground potential, while a gate electrode G which is a control electrode of the TFT 3854 is connected to the scanning line 3852. When the scanning line 3852 is at a high-level potential, the TFT 3854 enters a conducting state so that a picture signal which is carried by the signal line 3851 is written in the liquid crystal element 3853. When the scanning line 3852 is at a low-level potential, on the other hand, the TFT 3854 enters a cutoff state, to hold the picture signal which is written in the liquid crystal element 3853. A number of scanning lines 3852 are successively brought into high levels one by one, for example, for successively updating images displayed by this unit. A holding capacitance 3855 is connected to the liquid crystal element 3853 for supplementing electrostatic capacitance of the liquid crystal element 3853, so that an image signal which is written in the liquid crystal element 3853 is sufficiently held over a period up to next writing.
(Problem of Prior Art 25)
In relation to the TFT 3854 which is employed for the active matrix LCD, known are two types of TFTs including an amorphous silicon TFT and a polysilicon TFT employing amorphous Si polycrystalline Si (polysilicon) for active layers thereof respectively. The term "active layer" in relation to an element such as a transistor indicates a principal part of an element having semiconductor regions of different conductivity types and a junction therebetween for implementing behavior of a main current which is specific to the element.
&lt;1. Problems of Unit Employing Amorphous Silicon TFT&gt;
An active layer of an amorphous silicon TFT can be formed under a low temperature through chemical vapor deposition (CVD), for example. Therefore, the process maximum temperature for forming such an amorphous silicon TFT can be reduced below 400.degree. C. Thus, a transparent substrate to be provided with the TFT can be prepared from a low-priced glass substrate having a low withstand temperature. In other words, it is possible to fabricate an amorphous silicon TFT at a low cost.
On the other hand, such an amorphous silicon TFT having an active layer of amorphous Si disadvantageously has a small ON-state current, i.e., current upon conduction, since the active layer has low mobility of about 0.2 to 0.5 cm.sup.2 /(V.multidot.sec). Thus, it is difficult to obtain a high-contrast image having small flicker noise. Further, it is difficult to refine the element due to the small mobility, and hence it is difficult to densely arrange pixels. Consequently, it is difficult to obtain a high-definition image.
Since the active layer is made of amorphous Si, the so-called inverted stagger structure in which a gate electrode is positioned under a channel portion on the side of the transparent substrate is employed. Therefore, it is impossible to use the self-alignment technique of introducing an impurity through the gate electrode serving as a mask. Consequently, the TFT must inevitably be increased in size due to requirement for redundant design of about 5 .mu.m. This also leads to difficulty in provision of a high-definition image. Further, no excellent switching characteristics can be attained since parasitic capacitance between electrodes is increased due to the large-sized TFT, leading to deterioration of picture quality.
In addition, it is difficult to form a circuit for driving the amorphous silicon TFT on a single transparent substrate with the amorphous silicon TFT.
&lt;2. Problems of Unit Employing Polysilicon TFT&gt;
As compared with the amorphous silicon TFT having the aforementioned characteristics, a polysilicon TFT has the following characteristics: The polysilicon TFT has a large ON-state current, i.e., current upon conduction, since its active layer which is made of polycrystalline Si has high mobility of about 10 to 50 cm.sup.2 /(V.multidot.sec.). Therefore, a high-definition image having high contrast and small flicker noise can be easily obtained.
Since the active layer is made of polycrystalline Si, it is possible to employ such a structure that a gate electrode is positioned above a channel portion, i.e., on a side opposite to the transparent substrate. Thus, it is possible to employ the self-alignment structure of introducing an impurity through the gate electrode serving as a mask by preparing the gate electrode from polycrystalline Si, whereby no redundant design is required. This also contributes to a high-definition image. Further, a circuit for driving the polysilicon TFT can be easily formed on a single transparent substrate with the polysilicon TFT, whereby it is possible to advantageously form a miniature unit which is easy to handle.
However, the process maximum temperature of the polysilicon TFT exceeds 600.degree. C. so far as the same is fabricated by a conventional method in prior art 25, since it is necessary to form a polycrystalline Si thin film. Thus, an SiO.sub.2 (quartz) substrate having a high withstand temperature must be employed as the transparent substrate to be provided with the TFT and the like. Namely, the fabrication cost is increased in a unit employing the polysilicon TFT.
As hereinabove described, a unit employing an amorphous silicon TFT is inferior in picture quality although the same can be fabricated at a low cost, while the polysilicon TFT has a problem of a high fabrication cost although the same has excellent picture quality. In a unit employing a transistor whose active layer is made of single-crystalline Si in place of a polysilicon TFT, further, the fabrication cost is further increased although the picture quality is further improved, while the transistor cannot be provided in the form of a thin film but only a bulk transistor is available, and hence only a reflective LCD can be formed.
(Prior Art 26)
FIGS. 288 to 290 show a semiconductor device of a metal-insulator-semiconductor structure (MIS structure) according to Prior Art 26 employing a cladding system. In the cladding system, an oxide film 3902 is first grown on an Si substrate 3901 to form a first sample 3903 as shown in FIG. 288. Then another silicon wafer sample 3904 is clad thereon as shown in FIG. 289, and the as-clad silicon wafer sample 3904 is scraped to be reduced in thickness as shown in FIG. 290, thereby forming a transistor.
(Prior Art 27)
FIGS. 291 and 292 show a semiconductor device according to Prior Art 27 employing a SIMOX system. According to the SIMOX system, oxygen ions 3912 are ion-implanted in high concentration into a portion of a desired depth in an Si substrate 3911, as shown in FIG. 291. Thereafter a heat treatment is carried out to react the as-implanted oxygen ions 3912 with silicon which is contained in the Si substrate 3911 for forming an SiO.sub.2 film 3913 thereby completely dielectric-isolating an upper Si layer 3914 from a lower Si layer 3915 as shown in FIG. 292, so that a heterojunction transistor is formed on the upper surface.
(Prior Art 28)
FIG. 293 illustrates a step of forming a channel in a MOSFET of a general MIS structure or a metal-ferroelectric-semiconductor structure (MFS structure) for serving as an electrically writable/erasable non-volatile memory element according to Prior Art 28. In Prior Art 28, a LOCOS oxide film 3922 and a gate oxide film 3923 are first partially formed on an upper surface of an Si substrate 3921, as shown in FIG. 293. Then, an impurity such as B.sup.+ is ion-implanted into the Si substrate 3921 to form a channel, in order to control the threshold value of the MOSFET. At this time, the ion implantation is carried out through the gate oxide film 3923 in the state shown in FIG. 293, or the gate oxide film 3923 is wet-etched after the ion implantation to re-oxidize the Si substrate 3921 for forming a new gate oxide film.
FIG. 294 is a sectional view showing a step of diffusing/forming a source 3924 and a drain 3925 of the general MOSFET according to Prior Art 28. Ions as implanted are generally prepared from P.sup.+ or As.sup.+ for an NMOS transistor, or B.sup.+ or BF.sub.2.sup.+ for a PMOS transistor. Referring to FIG. 294, numeral 3926 denotes a gate.
FIG. 295 is a sectional view showing the general MOSFET according to Prior Art 28, in a state immediately after formation of an interconnection film 3927. This interconnection film 3927 is generally prepared from Al--Si or Al--Si--Cu. Referring to FIG. 295, numeral 3928 denotes an insulating film of PSG, BPSG or NSG. An Al alloy/TiN laminate structure may be employed in general, since sufficient reliability may not be attained by interconnection with a single Al alloy layer, following refinement of the device.
FIG. 296 is a sectional view showing the MOSFET of a two-layer interconnection film system according to Prior Art 28. Referring to FIG. 296, numeral 3929 denotes a second interconnection film of Al--Si or Al--Si--Cu, and numeral 3931 denotes an interlayer isolation film. The interlayer isolation film 3931, which is an underlayer for the second interconnection film 3929, is preferably flattened since the second interconnection film 3929 may be disconnected if this film 3931 has a step.
In order to flatten the interlayer isolation film 3931, an NSG, PSG or BPSG film or a multilayer oxide film thereof is first deposited by CVD and thereafter coated with a resist or SOG film by a spin coater, to fill up a step portion by a method such as sintering. Thereafter the as-coated material and the multilayer oxide film are simultaneously etched back to form the interlayer isolation film 3931. At this time, it is necessary to maintain the as-coated material and the multilayer oxide film at the same etching rates.
(Prior Art 29)
FIG. 297 is a sectional view showing an EPROM or a flash EPROM according to Prior Art 29. Referring to FIG. 297, numeral 3935 denotes an Si substrate, numeral 3936 denotes a source, numeral 3937 denotes a drain, numeral 3938 denotes a gate insulating film, numeral 3939 denotes a floating gate, numeral 3941 denotes an interlayer isolation film, and numeral 3942 denotes a control gate. The gate insulating film 3938 is formed by an oxide film which is prepared by thermally oxidizing the Si substrate 3935.
(Prior Art 30)
FIG. 298 is a sectional view showing a planar type NPN bipolar transistor according to Prior Art 30. Referring to FIG. 298, numeral 3945 denotes an n-type collector, numeral 3946 denotes a p-type base, and numeral 3947 denotes an n-type emitter. According to Prior Art 30, impurity ions of P or As are implanted into the n-type collector 3945 and the n-type emitter 3947 while impurity ions of B or BF.sub.2 are implanted into the p-type base 3946, and these impurities are diffused and activated by a heat treatment.
(Problem of Prior Art 26)
When the cladding system is employed as in Prior Art 26, however, the cost is increased since an operation for cladding two wafers with each other and scraping the same in high accuracy requires a considerable number of steps.
(Problem of Prior Art 27)
When the SIMOX system is employed as in Prior Art 27, on the other hand, considerably high energy ions are implanted and hence crystals are so significantly defected that it is difficult to attain excellent characteristics.
(Problem of Prior Art 28)
When the method of implanting ions through the gate oxide film 3923 in the state shown in FIG. 293 is employed in Prior Art 28, the film quality is deteriorated to easily cause a leakage current and dielectric breakdown due to passage of the high-energy ions through the gate oxide film 3923, although the threshold value can be controlled in high accuracy.
When the method of wet-etching the gate oxide film 3923 after the ion implantation for re-oxidizing the Si substrate 3921 and forming a new gate oxide film, on the other hand, it is difficult to accurately control the threshold value since an impurity (boron) which is implanted into Si is incorporated in the oxide film during the oxidation step, although a high-quality film can be obtained due to the new oxide film formed after the ion implantation.
Thus, it is difficult to improve the quality of the gate oxide film while accurately controlling the threshold value at the same time.
In Prior Art 28, further, impurity concentration generally exceeds 1.times.10.sup.20 cm.sup.-3 after the ion implantation for forming the source 3924 and the drain 3925, to cause a number of defects in the regions provided with the source 3924 and the drain 3925. A heat treatment (annealing) is required in order to recover the defects and activate the impurity. At this time, the impurity is disadvantageously diffused in Si due to the high-temperature treatment, and hence it is difficult to form refined shallow junction. According to Prior Art 28, the gate insulating film 3923 is designed to be about 100 .ANG. in thickness following refinement of the semiconductor device, while such a gate insulating film 3923 is rather inferior in reliability.
In the interconnection film according to Prior Art 28, sufficient reliability cannot be obtained by interconnection through a single Al alloy layer. In order to solve this problem, an Al alloy/TiN laminate structure is generally employed as hereinabove described. In this case, however, the fabrication steps are complicated to cause increase in cost.
Further, Prior Art 28 requires an oxide film of NS, PSG or BPSG and a coating material of resist or SOG provided on its upper surface for flattening the interlayer isolation film 3931 as shown in FIG. 296, while these materials must be maintained at the same etching rates. Thus, it takes time to adjust the materials.
(Problem of Prior Art 29)
In Prior Art 29, the gate insulating film 3923 is designed to be about 100 .ANG. in thickness following refinement of the semiconductor device similarly to Prior Art 28, while the gate insulating film 3923 must be improved in reliability in such a thin film forming technique. Thus, awaited is a technique for increasing the thickness of the gate insulating film 3923 and preventing carrier trap in the film interface and occurrence of a leakage current, thereby remarkably improving the insulating film withstand voltage. Particularly in the case of the EPROM or the flash EPROM according to Prior Art 29, the fabrication steps are complicated to increase the cost due to the four-layer structure of the gate portion as shown in FIG. 297.
(Problem of Prior Art 30)
In Prior Art 30, the size of the base 3946 is easily fluctuated due to large dispersion of its thickness D1, to extremely influence on the electric characteristics of the transistor. In the planar type device according to Prior Art 30, it is necessary to reduce the junction capacitance across the base and the collector as well as that across the emitter and the base in order to improve the switching speed of the transistor. However, it is difficult to reduce these junction capacitances since the emitter and the base are formed by diffusion, as hereinabove described. Further, it is also difficult to reduce the transistor size for a similar reason. FIG. 299 shows a polyemitter transistor structure, which is proposed in relation to a method for solving such a problem. In this structure, an emitter 3947 is formed by a polycrystalline film, so that the junction capacitance across the emitter 3947 and a base 3946 is set at a small value by forming an insulating film 3948. When such a method is employed, it is possible to form an extremely shallow diffusion region of the emitter 3947. However, refinement of the base 3946 is limited since the same is formed by a method similar to that in a planar type device, while the transistor characteristics may be further dispersed as compared with those of the planar type device in view of fabrication steps.